Next generation CMB experiments such as CMB-S4 aim at measuring the CMB lensing potential at sub-percent precision where most of the constraining power will come from CMB polarization. We investigate the prospects of achieving this goal in the presence of large-scale, diffuse galactic foreground emission by using non-Gaussian sky simulations and exploit multi-frequency information to clean those. We show that, while prior to foreground cleaning, cosmological parameter estimates from the contaminated lensing potential estimation can be significantly biased, these can be successfully mitigated by applying a parametric foreground cleaning approach. We further observe no significant additional bias in the delensed B-mode power spectrum after applying foreground cleaning and are therefore able to obtain an unbiased measurement of the tensor-to-scalar ratio, $r$, after delensing.
Comment: 26 pages, 13 figures, comments welcome